blob: 63e9ecae13103de18315f334cf553ac8031955db [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001#ifndef _ASM_PGALLOC_H
2#define _ASM_PGALLOC_H
3
4#include <linux/gfp.h>
5#include <linux/mm.h>
6#include <linux/threads.h>
7#include <asm/processor.h>
8#include <asm/fixmap.h>
9
10#include <asm/cache.h>
11
12/* Allocate the top level pgd (page directory)
13 *
14 * Here (for 64 bit kernels) we implement a Hybrid L2/L3 scheme: we
15 * allocate the first pmd adjacent to the pgd. This means that we can
16 * subtract a constant offset to get to it. The pmd and pgd sizes are
Helge Deller513e7ec2007-01-28 15:09:20 +010017 * arranged so that a single pmd covers 4GB (giving a full 64-bit
Linus Torvalds1da177e2005-04-16 15:20:36 -070018 * process access to 8TB) so our lookups are effectively L2 for the
19 * first 4GB of the kernel (i.e. for all ILP32 processes and all the
20 * kernel for machines with under 4GB of memory) */
21static inline pgd_t *pgd_alloc(struct mm_struct *mm)
22{
23 pgd_t *pgd = (pgd_t *)__get_free_pages(GFP_KERNEL,
24 PGD_ALLOC_ORDER);
25 pgd_t *actual_pgd = pgd;
26
27 if (likely(pgd != NULL)) {
28 memset(pgd, 0, PAGE_SIZE<<PGD_ALLOC_ORDER);
Helge Deller513e7ec2007-01-28 15:09:20 +010029#ifdef CONFIG_64BIT
Linus Torvalds1da177e2005-04-16 15:20:36 -070030 actual_pgd += PTRS_PER_PGD;
31 /* Populate first pmd with allocated memory. We mark it
32 * with PxD_FLAG_ATTACHED as a signal to the system that this
33 * pmd entry may not be cleared. */
34 __pgd_val_set(*actual_pgd, (PxD_FLAG_PRESENT |
35 PxD_FLAG_VALID |
36 PxD_FLAG_ATTACHED)
37 + (__u32)(__pa((unsigned long)pgd) >> PxD_VALUE_SHIFT));
38 /* The first pmd entry also is marked with _PAGE_GATEWAY as
39 * a signal that this pmd may not be freed */
40 __pgd_val_set(*pgd, PxD_FLAG_ATTACHED);
41#endif
42 }
43 return actual_pgd;
44}
45
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -080046static inline void pgd_free(struct mm_struct *mm, pgd_t *pgd)
Linus Torvalds1da177e2005-04-16 15:20:36 -070047{
Helge Deller513e7ec2007-01-28 15:09:20 +010048#ifdef CONFIG_64BIT
Linus Torvalds1da177e2005-04-16 15:20:36 -070049 pgd -= PTRS_PER_PGD;
50#endif
51 free_pages((unsigned long)pgd, PGD_ALLOC_ORDER);
52}
53
54#if PT_NLEVELS == 3
55
56/* Three Level Page Table Support for pmd's */
57
58static inline void pgd_populate(struct mm_struct *mm, pgd_t *pgd, pmd_t *pmd)
59{
60 __pgd_val_set(*pgd, (PxD_FLAG_PRESENT | PxD_FLAG_VALID) +
61 (__u32)(__pa((unsigned long)pmd) >> PxD_VALUE_SHIFT));
62}
63
64static inline pmd_t *pmd_alloc_one(struct mm_struct *mm, unsigned long address)
65{
66 pmd_t *pmd = (pmd_t *)__get_free_pages(GFP_KERNEL|__GFP_REPEAT,
67 PMD_ORDER);
68 if (pmd)
69 memset(pmd, 0, PAGE_SIZE<<PMD_ORDER);
70 return pmd;
71}
72
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -080073static inline void pmd_free(struct mm_struct *mm, pmd_t *pmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -070074{
Helge Deller513e7ec2007-01-28 15:09:20 +010075#ifdef CONFIG_64BIT
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 if(pmd_flag(*pmd) & PxD_FLAG_ATTACHED)
Mikulas Patocka0e0da482015-03-18 13:42:38 -040077 /*
78 * This is the permanent pmd attached to the pgd;
79 * cannot free it.
80 * Increment the counter to compensate for the decrement
81 * done by generic mm code.
82 */
83 mm_inc_nr_pmds(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 return;
85#endif
86 free_pages((unsigned long)pmd, PMD_ORDER);
87}
88
89#else
90
91/* Two Level Page Table Support for pmd's */
92
93/*
94 * allocating and freeing a pmd is trivial: the 1-entry pmd is
95 * inside the pgd, so has no extra memory associated with it.
96 */
97
98#define pmd_alloc_one(mm, addr) ({ BUG(); ((pmd_t *)2); })
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -080099#define pmd_free(mm, x) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100#define pgd_populate(mm, pmd, pte) BUG()
101
102#endif
103
104static inline void
105pmd_populate_kernel(struct mm_struct *mm, pmd_t *pmd, pte_t *pte)
106{
Helge Deller513e7ec2007-01-28 15:09:20 +0100107#ifdef CONFIG_64BIT
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 /* preserve the gateway marker if this is the beginning of
109 * the permanent pmd */
110 if(pmd_flag(*pmd) & PxD_FLAG_ATTACHED)
111 __pmd_val_set(*pmd, (PxD_FLAG_PRESENT |
112 PxD_FLAG_VALID |
113 PxD_FLAG_ATTACHED)
114 + (__u32)(__pa((unsigned long)pte) >> PxD_VALUE_SHIFT));
115 else
116#endif
117 __pmd_val_set(*pmd, (PxD_FLAG_PRESENT | PxD_FLAG_VALID)
118 + (__u32)(__pa((unsigned long)pte) >> PxD_VALUE_SHIFT));
119}
120
121#define pmd_populate(mm, pmd, pte_page) \
122 pmd_populate_kernel(mm, pmd, page_address(pte_page))
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800123#define pmd_pgtable(pmd) pmd_page(pmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800125static inline pgtable_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126pte_alloc_one(struct mm_struct *mm, unsigned long address)
127{
128 struct page *page = alloc_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
Kirill A. Shutemovbc166402013-11-14 14:31:37 -0800129 if (!page)
130 return NULL;
131 if (!pgtable_page_ctor(page)) {
132 __free_page(page);
133 return NULL;
134 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 return page;
136}
137
138static inline pte_t *
139pte_alloc_one_kernel(struct mm_struct *mm, unsigned long addr)
140{
141 pte_t *pte = (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
142 return pte;
143}
144
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -0800145static inline void pte_free_kernel(struct mm_struct *mm, pte_t *pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146{
147 free_page((unsigned long)pte);
148}
149
Kyle McMartin9aa150b2008-02-18 14:16:26 -0800150static inline void pte_free(struct mm_struct *mm, struct page *pte)
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800151{
152 pgtable_page_dtor(pte);
Kyle McMartin9aa150b2008-02-18 14:16:26 -0800153 pte_free_kernel(mm, page_address(pte));
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800154}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156#define check_pgt_cache() do { } while (0)
157
158#endif